Advances and Limits to applying Agent Based Models in Water Risk Modelling

A particular feature of coupled human-water systems is that they are inherently complex, characterized by non-linearities, feedbacks, thresholds, and heterogeneity. Capturing these complexities in flood and drought risk models is a challenge. Therefore, existing risk models often include human responses to changes in the water system as scenarios that are static over time and space. However, in reality these responses are highly dynamic processes. Agent-based models (ABMs) are an emerging class of models to simulate such dynamic feedbacks between the hydrological and human systems. In ABMs, agents in water management (e.g. farmers, flood managers, insurers, etc.) can observe, learn from, move around, and influence (and are influenced by) the flood / drought risk they face, and can make decisions about the implementation of different adaptation actions. ABMs capture the temporal dynamics of risk by aggregating the results of numerous actions by individual agents, while they also allow to assess risk on an individual agent level. This presentation will address the recent advances and challenges in applying ABMs in both flood and drought risk management. For example, despite the availability of empirical data on water management from surveys, the calibration of decision rules in ABMs is hampered by the lack of longitudinal empirical data on individual risk reduction behaviour (e.g. responses to floods and droughts). In addition, we will show how to apply existing behavioural theories from economics and psychology to parameterize ABMs, which help to refine new surveys for collecting water management data. Furthermore, the validation of coupled hydro-ABMs is challenging as for flood risk, relatively few data points on flood events and losses are available. For drought ABMs, there seems better scope for overall validation because of the availability of drought impact data (e.g. yields) as compared to flood losses.